An injection neutral line protection system based on multi-dimensional feature fusion

Abstract

The application discloses an injection type neutral line protection system based on multi-dimensional feature fusion, relates to the technical field of neutral line protection of power distribution systems, and comprises a power grid state acquisition unit, a physical link detection unit and a central control unit. The power grid state acquisition unit is used for acquiring electrical operation data and extracting a passive feature vector reflecting the power frequency state of the power grid. The physical link detection unit injects a non-power frequency detection signal into the neutral line circuit and extracts an active feature vector reflecting the physical connection state of the neutral line based on a response signal. The central control unit performs attribute judgment on electrical abnormalities in the passive feature vector according to the impedance characteristics of the active feature vector, realizes accurate qualitative fault and protection instruction output, and effectively solves the blind area problem of traditional passive monitoring in the detection of hidden faults such as high-resistance contact, harmonic interference and critical section through active and passive feature fusion, and significantly improves the fault identification accuracy, early warning capability and system reliability.

Description

Technical Field

[0001] This invention relates to the field of neutral line protection technology in power distribution systems, specifically to an injection-type neutral line protection system based on multi-dimensional feature fusion. Background Technology

[0002] Currently, low-voltage distribution networks widely adopt three-phase four-wire power supply. With the neutral line intact, even if the three-phase load is unbalanced, the voltage of each phase can be kept stable. However, once the neutral line breaks (breaks zero), it will cause the neutral point potential to drift, leading to "burn-out" or equipment undervoltage shutdown accidents.

[0003] Existing neutral-line (N) protection technologies primarily rely on a "passive monitoring method," which involves collecting three-phase voltage, current, zero-sequence current, and harmonic components from the circuit and predicting faults using threshold judgments or AI models. However, this method has significant technical blind spots, accounting for approximately 32% of fault scenarios, including:

[0004] Misjudgment of high-resistance contact: When the resistance of the N-line connector increases due to corrosion or loosening but it is not completely disconnected, the voltage / current characteristics are not obviously abnormal under light load or three-phase balanced conditions, and the existing model misjudges it as "safe".

[0005] Harmonic interference: When there are many nonlinear loads (such as frequency converters and LEDs), the N line itself has a large third harmonic current. Existing models have difficulty distinguishing between "load characteristics" and "abnormalities caused by increased line impedance", which can easily lead to missed or false judgments.

[0006] Critical cross section: When the wire diameter is reduced due to theft or construction damage, passive parameters cannot detect the vulnerability of the physical link before a failure occurs. Summary of the Invention

[0007] The purpose of this invention is to provide an injection-type neutral line protection system based on multi-dimensional feature fusion. This system captures the power grid frequency status through a passive channel and obtains the physical link status through an active channel. The two channels interact deeply at the feature engineering layer to eliminate the detection blind spot of a single dimension and solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an injection-type neutral line protection system based on multi-dimensional feature fusion, comprising:

[0009] The power grid status acquisition unit is configured to acquire electrical operation data of the power distribution system and extract passive feature vectors that reflect the power frequency status of the power grid.

[0010] The physical link detection unit is configured to inject a non-power frequency detection signal into the neutral line loop and extract an active feature vector reflecting the physical connection status of the neutral line based on the response signal fed back from the detection signal.

[0011] The central control unit performs multi-dimensional feature fusion based on the passive feature vector and the active feature vector to construct a prediction model. The prediction model includes a multi-attribute verification layer. The multi-attribute verification layer performs fault attribute qualitative analysis on the electrical anomaly indicated by the passive feature vector based on the physical impedance characteristics of the active feature vector as a benchmark, defines the real neutral line fault, and outputs the corresponding protection command.

[0012] Preferably, the physical link detection unit includes a signal injection module and a signal extraction module;

[0013] The signal extraction module is configured to separate the response signal component with the same injection frequency from the mixed signal containing strong power frequency interference based on frequency domain separation technology, and obtain the impedance amplitude parameter in the active feature vector based on the complex relationship between the voltage and current of the separated response signal component.

[0014] Preferably, the attribute determination logic includes:

[0015] Current characteristic identification logic: In response to the current parameter exceeding the limit in the passive feature vector, the impedance amplitude parameter in the active feature vector is used to determine the source of the current anomaly, so as to distinguish between nonlinear superposition on the load side and high impedance fault on the line side.

[0016] Dynamic sensitivity adjustment logic: Real-time monitoring of impedance stability parameters in the active feature vector. When impedance instability is detected to exceed the preset level, the fault trigger threshold or weight of voltage parameters in the passive feature vector is automatically adjusted to identify potential contact hazards under low voltage drift conditions.

[0017] Fault attribution logic: In response to the abnormal three-phase imbalance parameters in the passive eigenvector, the impedance amplitude parameter in the active eigenvector is used to verify the physical continuity of the neutral line, so as to distinguish between quality problems on the power supply side and open circuit faults on the line side.

[0018] Preferably, the specific configuration of the dynamic sensitivity adjustment logic includes: when the impedance stability parameter characterizes the physical connection as loose or micro-arc characteristics, lowering the judgment threshold for neutral-to-ground voltage alarm to achieve early warning;

[0019] The impedance stability parameter in the active feature vector is a statistical indicator used to characterize the degree of dispersion or the severity of fluctuation of the impedance amplitude parameter within a certain time window.

[0020] Preferably, the current property discrimination logic is specifically configured as follows:

[0021] If the current parameter exceeds the limit and the impedance amplitude parameter is in the normal low impedance range, it is determined to be a non-faulty harmonic overload condition.

[0022] If the current parameter exceeds the limit and the impedance amplitude parameter is significantly higher than the reference value, it is determined to be a neutral line fault state and a protection action command is generated.

[0023] Preferably, the specific configuration of the fault attribution logic includes:

[0024] If the three-phase imbalance parameters are abnormal but the impedance amplitude parameters show that the physical link is intact, the abnormality is determined to be caused by external power input, and the zero-break protection is blocked.

[0025] If the three-phase imbalance parameters are abnormal and the impedance amplitude parameters show that the physical link is broken or the impedance is high, the abnormality is determined to be a neutral line breakage fault.

[0026] Preferably, the prediction model further includes a state classification layer, which is used to input the passive feature vector and the active feature vector after concatenation and normalization, and output the health status category of the current neutral line. The health status category includes healthy state, poor contact potential state, corrosion and aging state, and wire breakage fault state.

[0027] Preferably, the passive feature vector includes at least one of the following: time-domain features reflecting steady-state amplitude, frequency-domain features reflecting waveform distortion, and symmetric component features reflecting the degree of three-phase asymmetry.

[0028] Preferably, the non-power frequency detection signal injected by the physical link detection unit is an AC signal or pulse signal with a frequency different from the fundamental frequency of the power grid and its main harmonic frequencies.

[0029] Preferably, the frequency domain separation technique includes at least one of digital quadrature demodulation, discrete Fourier transform, or bandpass filtering, used to obtain the in-phase and quadrature components of the response signal.

[0030] In summary, the beneficial effects of this invention are:

[0031] This invention constructs an "active-passive fusion sensing model," innovatively introducing active detection signals to acquire real-time impedance and other characteristics of the neutral line's physical link, building upon traditional passive monitoring of the power grid's power frequency status. The active features are then used to decouple and correct the passive features. Through the logical interaction of multi-dimensional features—such as using active impedance to verify current properties, using impedance fluctuations to amplify voltage drift sensitivity, and using loop reactance to correct imbalance attribution—this system effectively eliminates the blind spots of single passive monitoring. It achieves accurate identification and early warning of hidden faults such as high-resistance contacts, harmonic interference, and critical cross-sections, significantly reducing false alarms and missed alarms while greatly advancing fault detection time, thereby improving the reliability and intelligence level of neutral line lifecycle health management. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a schematic diagram of the process framework structure of an injection-type neutral line protection system based on multi-dimensional feature fusion according to the present invention.

[0034] Figure 2 This is a schematic diagram of signal comparison in an injection-type neutral line protection system based on multi-dimensional feature fusion according to the present invention.

[0035] Figure 3 This is a flowchart illustrating auxiliary logic one in an injection-type neutral line protection system based on multi-dimensional feature fusion according to the present invention.

[0036] Figure 4 This diagram illustrates a comparative analysis of active and passive signals in an injection-type neutral line protection system based on multi-dimensional feature fusion, as described in this invention. Detailed Implementation

[0037] The present invention will now be described in further detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. These drawings are simplified schematic diagrams, which are only used to illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0038] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate several embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0039] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0040] Any feature disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by other equivalent or similar features for a similar purpose, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0041] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; they can refer to the internal communication of at least two elements or the interaction relationship of at least two elements, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0042] Please see Figures 1-4 This invention provides an embodiment of an injection-type neutral line protection system based on multi-dimensional feature fusion. The core of this system lies in constructing an "active-passive fusion sensing model": Based on the traditional passive layer that collects current / voltage characteristics, an active layer is constructed by introducing a specific frequency active detection signal. By injecting pulsed DC or alternating frequency AC signals that do not interfere with the power grid frequency, the characteristic data of the neutral line circuit are directly measured. These characteristics are then combined with the passive layer characteristics to construct a predictive model, achieving precise health management of the neutral line throughout its entire lifecycle. Specifically, this includes the following parts:

[0043] Part 1: Power frequency data acquisition, analysis and feature extraction on the passive side of the power grid;

[0044] The goal is to comprehensively capture the power grid's operating status, identify imbalances and voltage drift caused by zero-point disconnection, and the harmonic characteristics of nonlinear loads;

[0045] 1. Data collection strategy:

[0046] Sampling frequency: set to f≥6.4kHz, that is, 128 points are sampled per power frequency cycle to meet the analysis requirements of at least 63 harmonics;

[0047] Data collected includes: three-phase voltages (Ua, Ub, Uc), three-phase currents (ia, ib, ic), neutral current (in), and neutral-to-ground voltage (Un).

[0048] Passive Feature Extraction: We will extract three types of features to form a passive feature vector Vp, specifically including:

[0049] A. Time-domain statistical characteristics, used to capture transient fluctuations and overall amplitude levels, reflecting steady-state amplitude.

[0050] Effective value (RMS): Calculates the RMS value of each phase voltage and current.

[0051] Zero-sequence voltage (U0): Calculate the vector sum of Ua+Ub+Uc. Under normal circumstances, it is close to 0. When the N line breaks, this value increases significantly, which is the most direct evidence of zero-sequence failure.

[0052] Neutral line to ground voltage (Un): This is a direct measurement of the potential difference between the N line and the PE line, reflecting the voltage drop on the N line. If the current is small but the voltage drop is large, it indicates that the line impedance is large.

[0053] B. Frequency domain characteristics: Waveform distortion is analyzed using Fast Fourier Transform (FFT) to reflect waveform distortion.

[0054] Total Harmonic Distortion (THD): Especially the THD of in.

[0055] 3rd harmonic percentage (HR3): Calculation .

[0056] To distinguish between a "fault" and a "harmonic overload," if the current in the neutral line is large and the proportion of the third harmonic is extremely high, it is usually due to the load rather than a line break.

[0057] C. Symmetrical component characteristics, reflecting the degree of three-phase asymmetry, specifically the three-phase imbalance. ):

[0058] ;

[0059] in It is a negative sequence current. It is the positive sequence current, used to assess the degree of load asymmetry and help determine the source of the neutral line current.

[0060] Part Two: Analysis and Feature Extraction of Active-Side Injected Signals;

[0061] Extracting weak injection signals from strong power frequency interference and acquiring the response voltage of the injection signal loop ( ) and response current ( ), calculate the true impedance of the physical circuit, and identify "hidden" faults.

[0062] 1. Signal extraction: Frequency domain separation technology is used to separate the response signal component with the same injection frequency from the mixed signal containing strong power frequency interference. The frequency domain separation technology includes at least one of digital quadrature demodulation, discrete Fourier transform or bandpass filtering, which is used to obtain the in-phase component and quadrature component of the response signal.

[0063] Due to significant power frequency interference, simple filtering is insufficient. It is recommended to use digital quadrature demodulation to extract specific frequencies, such as... The signal specifically includes the following steps:

[0064] Step S1: Constructing the reference signal

[0065] Two orthogonal reference signals are generated inside the MCU:

[0066] ;

[0067] ;

[0068] Step S2: Quadrature mixing

[0069] The acquired noisy feedback voltage and current Multiply by the reference signal respectively:

[0070] ;

[0071] ;

[0072] Step S3: Low-pass filtering

[0073] By using a digital low-pass filter with an extremely low cutoff frequency, such as 5Hz, all AC components are filtered out, leaving only the DC component, thus obtaining the in-phase component of the voltage. and orthogonal components Similarly, the current components are obtained. and .

[0074] 2. Active feature extraction, constructing an active feature vector Va:

[0075] A. Physical impedance characteristics;

[0076] circuit resistance ( ): Calculate the real part of the loop resistance using complex number operations:

[0077] ;

[0078] This is the most crucial physical indicator for determining whether the N-line is corroded, has poor contact, or is broken.

[0079] Circuit reactance ( Changes in inductance can reflect changes in the circuit path or wire theft, such as a shortened wire length.

[0080] B. Stability characteristics;

[0081] Impedance variance : Calculate the loop resistance in the past second ( The variance of the fluctuation.

[0082] Key characteristics: poor contact (loosening) often manifests as violent fluctuations in resistance within a small range (arc noise), while corrosion increases monotonically and slowly. Large variance indicates that the screw is loose.

[0083] C. Signal-to-noise ratio (SNR) characteristics

[0084] Extracting the ratio of signal energy at the injection frequency to the noise floor energy at adjacent frequencies, a sharp drop in SNR may indicate strong arc discharge interference in the line.

[0085] Part Three: Auxiliary Interaction Logic and Multidimensional Feature Fusion

[0086] For the auxiliary interaction logic, the active features extracted in the second part constitute the "decoupling" and "benchmark correction" of the passive features, specifically including the following auxiliary logic:

[0087] Auxiliary Logic 1: Current Characteristic Identification. In response to the current parameter exceeding limits in the passive eigenvector, the source of the current anomaly is determined using the impedance amplitude parameter in the active eigenvector, to differentiate between nonlinear superposition on the load side and high-impedance faults on the line side. Specifically:

[0088] Utilizing loop resistance ( The neutral line current (in) is "qualitatively characterized" to solve the problem of not being able to distinguish between "nonlinear load harmonic overload (normal)" and "current abnormality (fault) caused by high impedance fault" when the neutral line current (in) rises.

[0089] If there is only a single passive characteristic, its limitations are as follows:

[0090] When the system detects a large neutral current (in), the passive model cannot make a judgment:

[0091] Situation A: This is because a large number of LEDs / inverters in the load generate third harmonics. This is a normal electrical phenomenon and does not require tripping; only capacity expansion is needed.

[0092] Situation B: The increased contact resistance of the neutral line causes abnormal current under the same voltage deviation, which is a fault and requires tripping.

[0093] Implementation logic: Establish an attribute validation gate; input: passive features Active features .

[0094] like _th (current exceeded) and (Impedance maintained at a low reference level):

[0095] The system determined that there was a "source-side harmonic overload", blocked the trip signal, and only issued an alarm that "suggests capacity expansion or harmonic mitigation".

[0096] like _th (current exceeded) and (Impedance significantly increased):

[0097] The system determined that there was a "grid-side line fault" and immediately triggered the circuit breaker to trip.

[0098] By using the active impedance as the "true value", false positives caused by load characteristics are eliminated.

[0099] Auxiliary Logic 2: Dynamic Sensitivity Adjustment. This feature monitors the impedance stability parameter in the active feature vector in real time. When impedance instability exceeds a preset level, it automatically adjusts the fault trigger threshold or weight of the voltage parameter in the passive feature vector to identify potential contact hazards under low-voltage drift conditions. Details are as follows:

[0100] use (Active) Significantly enhances the voltage drift sensitivity of the neutral line to ground voltage Un (passive) to achieve early warning;

[0101] This addresses the blind spot where changes in Un are subtle during the initial stages of poor contact or under light load, making it undetectable by the passive model.

[0102] In passive monitoring, the neutral-to-ground voltage Un will only increase when poor contact develops to a certain extent, such as when the resistance is very high and a large load current flows through the line.

[0103] In the early stages of light load or poor contact, the screw has just loosened, occasionally sparks occur, and Un hardly changes; the passive model is "invisible," while the active feature... Its auxiliary function is as a "magnifying glass":

[0104] It is the variance (fluctuation rate) of impedance. The most significant physical characteristic of mechanical loosening is that the contact resistance fluctuates drastically within a small range.

[0105] Auxiliary logic: Establish a dynamic weighting mechanism, input: passive feature Un and active feature ;

[0106] monitor Under normal connection conditions, the impedance variance is extremely small, essentially approaching zero.

[0107] When detected When the silence threshold is exceeded and high-frequency jitter occurs, indicating microscopic arcing or loosening, the system automatically lowers the alarm threshold for Un, for example, from 24V to 5V, or increases the input weight of Un in the model.

[0108] Even if Un has not yet reached the traditional danger level at this time, combined with high Despite its inherent characteristics, the system still outputs a "contact hazard warning," giving it the ability to sense faults in zero-current / micro-current conditions and significantly advancing the time of fault detection.

[0109] Auxiliary Logic 3: Fault Attribution. Responding to abnormal three-phase imbalance parameters in the passive eigenvector, the impedance amplitude parameter in the active eigenvector is used to verify the physical continuity of the neutral line, thus distinguishing between power supply-side quality issues and line-side open circuit faults. Specifically:

[0110] use (Active) Correction The "attribution direction" of (passive three-phase unbalance) solves the problem of not being able to distinguish whether the problem is a phase loss in the upstream power supply or a broken neutral wire in the unit when the three-phase voltage is unbalanced.

[0111] The limitation of passive features alone is:

[0112] When a severe imbalance occurs in the three-phase voltage, such as low voltage in phase A and high voltage in phase B, the passive model will typically trigger an alarm. However, this could be due to:

[0113] Reason A: The upstream power supply itself is missing a phase or is unbalanced (source-side problem).

[0114] Reason B: Neutral point drift caused by a broken neutral line (net-side problem).

[0115] Traditional passive protection systems have difficulty distinguishing between these two types of faults and are prone to misjudging them as zero-break faults when the source-side voltage fluctuates.

[0116] Active features The auxiliary role (as "path verification") is to compare the theoretically calculated "zero offset" with the actual observed value in real time.

[0117] Auxiliary logic: Build a causal attribution tree. Input: passive features. (Three-phase imbalance); Active characteristics ;

[0118] Path A: Abnormal (High) + Normal (low).

[0119] Inference: The N-line physical link is intact; the voltage imbalance is caused by external input.

[0120] The problem was identified as an "external power supply quality issue," and undervoltage / overvoltage protection logic was executed, but no "zero-off fault" was reported.

[0121] Path B: Abnormal (High) + Abnormal (high / infinity).

[0122] Inference: The physical break of the N-line caused the neutral point to drift.

[0123] The fault is identified as a "zero-out fault," and the N-line protection logic is executed.

[0124] Clearly define the boundaries of fault responsibility to prevent the machine from misjudging the N-line fault due to fluctuations in the upstream power grid.

[0125] After the above logical processing, multi-dimensional feature fusion is performed to construct a prediction model:

[0126] Feature concatenation and normalization are used to construct a fused vector:

[0127] ;

[0128] Model selection: Random forest or LSTM network;

[0129] Training label design (including hidden faults):

[0130] Label 0 (Health): in any case. Low, Low.

[0131] Label 1 (Poor Contact): in is small, un is normal, but... High coverage, traditional blind spots.

[0132] Label 2 (Corrosion Aging): Not exceeded but With a positive slope, traditional blind spots are covered.

[0133] Label 3 (zero break): High, zero-sequence voltage (U0) High

[0134] The integration resulted in the following advantages:

[0135] Mutual verification mechanism: Active impedance characteristics As a "truth value," the passive current characteristics are verified. For example, when the passive characteristics show a voltage deviation, but the active characteristics show an extremely low loop impedance, the model will tend to judge it as a "source power quality problem" rather than a "line break," thereby avoiding false tripping.

[0136] Blind zone coverage: During the no-load moment when the load current is 0, all passive features become ineffective (all values ​​are 0). At this time, the model weights automatically tilt towards the active features, relying on... It can still identify the disconnection state from the input vector.

[0137] Furthermore, the present invention also provides the following specific operational examples:

[0138] The scenario involves a commercial office building's power distribution system, employing a three-phase four-wire power supply. Loads include non-linear loads such as LED lighting, air conditioning inverters, and computer servers. The neutral wire, located in a distribution box on a certain floor, has experienced increased contact resistance due to long-term corrosion (from approximately 0.1Ω to 5Ω), but it has not yet been completely disconnected.

[0139] The system operation process is as follows:

[0140] Step 1: Data Acquisition and Feature Extraction;

[0141] Passive feature acquisition (power grid status acquisition unit);

[0142] Sampling frequency: 6.4kHz;

[0143] Data collected: three-phase voltage, current, neutral current, and neutral-to-ground voltage;

[0144] Feature extraction:

[0145] Time domain: Neutral current RMS value In = 28A (slightly exceeds the threshold), neutral voltage to ground Un = 3V (below the alarm threshold of 24V);

[0146] Frequency domain: The third harmonic of the neutral current accounts for 45%, and the total harmonic distortion (THD) is 48%.

[0147] Symmetrical component: Three-phase imbalance =8% (mild abnormality);

[0148] Active feature acquisition (physical link detection unit):

[0149] Injected signal: 125Hz AC signal (non-power frequency and its harmonics);

[0150] Extracting the response signal: separating it through digital quadrature demodulation ( ) and response current ( ).

[0151] Calculate active features:

[0152] loop resistance =4.8Ω (significantly higher than the benchmark value of 0.1Ω);

[0153] Impedance variance =0.6 (significant fluctuations indicate unstable contact);

[0154] Signal-to-noise ratio (SNR) = 15dB (acceptable, no obvious arc interference);

[0155] Step 2: Feature fusion and logical judgment (central control unit)

[0156] Input fusion vector:

[0157] ;

[0158] The data is then input into the prediction model after Z-Score normalization.

[0159] Logical judgment process:

[0160] Current property identification logic:

[0161] Passive features: In > threshold, THD is high.

[0162] Active features: > .

[0163] Judgment: The current over-limit is not simply a harmonic overload, but a high-resistance fault in the line.

[0164] Output: Trigger "Neutral line high resistance fault" alarm, prepare trip command.

[0165] Dynamic sensitivity adjustment logic:

[0166] Active features: The threshold indicates that there is microscopic loosening in the contact.

[0167] The system automatically lowers the Un alarm threshold from 24V to 5V.

[0168] Currently, Un=3V, which has not reached the new threshold, but combined with high... It outputs a "potential for poor contact" warning.

[0169] Fault attribution logic:

[0170] Passive characteristics: =8% (slight imbalance).

[0171] Active features: High, but not infinite (not completely disconnected).

[0172] Judgment: The imbalance is mainly caused by the high resistance of the neutral line, rather than by the external power supply.

[0173] Output: Lock out "zero-loss protection" and start "high-resistance fault handling process".

[0174] Step 3: Status Classification and Protection Output

[0175] Output of the state classification layer of the prediction model:

[0176] Health status category: Corrosion and aging state (Label 2);

[0177] in accordance with: The upward trend is obvious. High, Un did not increase significantly but there are potential risks.

[0178] A warning message was sent to the maintenance platform: "Neutron contact resistance is abnormally high, indicating a risk of corrosion and aging." Planned maintenance is recommended to prevent it from developing into a wire breakage fault. If the value continues to rise above the safety threshold, trip protection will be activated.

[0179] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any variations or substitutions conceived without inventive effort should be included within the scope of protection of the invention. Therefore, the scope of protection of the invention should be determined by the scope defined in the claims.

Claims

1. An injection-type neutral line protection system based on multi-dimensional feature fusion, characterized in that, include: The power grid status acquisition unit is configured to acquire electrical operation data of the power distribution system and extract passive feature vectors that reflect the power frequency status of the power grid. The physical link detection unit is configured to inject a non-power frequency detection signal into the neutral line loop and extract an active feature vector reflecting the physical connection status of the neutral line based on the response signal fed back from the detection signal. The central control unit performs multi-dimensional feature fusion based on the passive feature vector and the active feature vector to construct a prediction model. The prediction model includes a multi-attribute verification layer. This layer uses the physical impedance characteristics of the active feature vector as a benchmark to perform attribute judgment logic on the electrical anomalies indicated by the passive feature vector, qualitatively defining the actual neutral line fault and outputting the corresponding protection command. The attribute judgment logic includes: Current characteristic identification logic: In response to the current parameter exceeding the limit in the passive feature vector, the impedance amplitude parameter in the active feature vector is used to determine the source of the current anomaly, so as to distinguish between nonlinear superposition on the load side and high impedance fault on the line side. Dynamic sensitivity adjustment logic: Real-time monitoring of impedance stability parameters in the active feature vector. When impedance instability is detected to exceed the preset level, the fault trigger threshold or weight of voltage parameters in the passive feature vector is automatically adjusted to identify potential contact hazards under low voltage drift conditions. Fault attribution logic: In response to the abnormal three-phase imbalance parameters in the passive eigenvector, the impedance amplitude parameter in the active eigenvector is used to verify the physical continuity of the neutral line, so as to distinguish between quality problems on the power supply side and open circuit faults on the line side.

2. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 1, characterized in that: The physical link detection unit includes a signal injection module and a signal extraction module; The signal extraction module is configured to separate the response signal component with the same injection frequency from the mixed signal containing strong power frequency interference based on frequency domain separation technology, and obtain the impedance amplitude parameter in the active feature vector based on the complex relationship between the voltage and current of the separated response signal component.

3. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 2, characterized in that: The specific configuration of the dynamic sensitivity adjustment logic includes: when the impedance stability parameter characterizes the physical connection as loose or micro-arc characteristics, lowering the judgment threshold for neutral-to-ground voltage alarm to achieve early warning; The impedance stability parameter in the active feature vector is a statistical indicator used to characterize the degree of dispersion or volatility of the impedance amplitude parameter within a certain time window.

4. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 3, characterized in that: The specific configuration of the current property discrimination logic includes: If the current parameter exceeds the limit and the impedance amplitude parameter is in the normal low impedance range, it is determined to be a non-faulty harmonic overload condition. If the current parameter exceeds the limit and the impedance amplitude parameter is significantly higher than the reference value, it is determined to be a neutral line fault state and a protection action command is generated.

5. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 4, characterized in that: The specific configuration of the fault attribution logic includes: If the three-phase imbalance parameters are abnormal but the impedance amplitude parameters show that the physical link is intact, the abnormality is determined to be caused by external power input, and the zero-break protection is blocked. If the three-phase imbalance parameters are abnormal and the impedance amplitude parameters show that the physical link is broken or the impedance is high, the abnormality is determined to be a neutral line breakage fault.

6. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 1, characterized in that: The prediction model also includes a state classification layer, which is used to input the passive feature vector and the active feature vector after concatenation and normalization, and output the health status category of the current neutral line. The health status category includes healthy state, poor contact potential state, corrosion and aging state, and wire breakage fault state.

7. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 6, characterized in that: The passive feature vector includes at least one of the following: time-domain features reflecting steady-state amplitude, frequency-domain features reflecting waveform distortion, and symmetric component features reflecting the degree of three-phase asymmetry.

8. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 1, characterized in that: The non-power frequency detection signal injected by the physical link detection unit is an AC signal or pulse signal with a frequency different from the fundamental frequency of the power grid and its main harmonic frequencies.

9. The injection-type neutral line protection system based on multi-dimensional feature fusion according to claim 2, characterized in that: The frequency domain separation technique includes at least one of digital quadrature demodulation, discrete Fourier transform, or bandpass filtering, used to obtain the in-phase and quadrature components of the response signal.

Images